ai safety
open source
+10
Jun 30, 2026
•
3 min read
The U.S. gated its newest AI models. A Chinese lab gave one away.
Plus: agents moving from scoring the loan to making the call, and a malicious AI tool that cleared review and reached 26,000 users.
model risk
+10
Jun 24, 2026
•
7 min read
Treat your AI agent like an inside threat
Plus: Lloyds aims AI agents at the moment of payment, an $88-a-week kit that clones trusted-brand sites, and Santander open-sources its AI guardrails.
open source
+10
May 28, 2026
•
4 min read
A 3-minute mortgage. How?
Plus: BNP's open-model bet, the 3% controls gap, and a nice conversation starter.
